Dictation apps
Juno vs FluidVoice
The best free dictation app on the Mac, and the only one here that also acts. Command Mode fires the action you mapped. Juno works out the steps you did not.
FluidVoice deserves the attention it gets. It is free, it is GPLv3, transcription runs on your Mac, it carries a roster of local models most paid apps do not offer, and it is quick enough that the developer puts a sub hundred millisecond figure on the site. Eleven thousand GitHub stars is not an accident.
It is also the one app on this page that does not simply stop at the text box. Command Mode will launch apps, run Shortcuts, fire system actions and execute terminal commands by voice, which is more than any other dictation app here attempts.
The catch is the word mapped. Command Mode does what you configured it to do, and it is not much use until you have done that configuring. It is a voice driven launcher, so it belongs to the same family as Shortcuts and Keyboard Maestro. Juno has no mapping step because it has no map. It looks at what is actually on the screen and works out the steps, which is why it can operate an app nobody wrote a command for.
Side by side
| Juno | FluidVoice | |
|---|---|---|
| Voice to text at your cursor | Yes | Yes |
| Transcription runs locally | Yes | Yes |
| Speech models offered | Whisper, several sizes | Whisper, Parakeet, Nemotron, Cohere, Apple Speech |
| Acting on what you said | Any app, worked out from the screen | Actions you mapped in advance |
| Setup before it can act | None | You configure the command mappings |
| Works in apps with no automation support | Yes, it reads the screen | No |
| Multi step tasks across apps | Yes, from one sentence | Only if you scripted the chain |
| Wake word | Yes, and you choose it | No, global hotkey |
| Price | Free | Free |
| Cost per task | Model usage for agent work, dictation is free | None |
| Licence | FSL-1.1-MIT, becomes MIT | GPLv3 |
| macOS required | 14 Sonoma or later | 15 Sequoia or later |
| Other platforms | None | Windows pre-build, iPhone coming |
Voice to text at your cursor
Juno
YesFluidVoice
YesTranscription runs locally
Juno
YesFluidVoice
YesSpeech models offered
Juno
Whisper, several sizesFluidVoice
Whisper, Parakeet, Nemotron, Cohere, Apple SpeechActing on what you said
Juno
Any app, worked out from the screenFluidVoice
Actions you mapped in advanceSetup before it can act
Juno
NoneFluidVoice
You configure the command mappingsWorks in apps with no automation support
Juno
Yes, it reads the screenFluidVoice
NoMulti step tasks across apps
Juno
Yes, from one sentenceFluidVoice
Only if you scripted the chainWake word
Juno
Yes, and you choose itFluidVoice
No, global hotkeyPrice
Juno
FreeFluidVoice
FreeCost per task
Juno
Model usage for agent work, dictation is freeFluidVoice
NoneLicence
Juno
FSL-1.1-MIT, becomes MITFluidVoice
GPLv3macOS required
Juno
14 Sonoma or laterFluidVoice
15 Sequoia or laterOther platforms
Juno
NoneFluidVoice
Windows pre-build, iPhone coming
Where FluidVoice is the better pick
- Speed. It is built around latency and offers models chosen for it, and on pure dictation it is likely faster than Juno today.
- The licence. GPLv3 now beats a licence that becomes MIT later, and for some people that difference decides it.
- It is free all the way down. Juno's dictation costs nothing either, but agent tasks run on a model you pay for. FluidVoice never sends you a bill.
- A wider choice of local speech models, including Parakeet and Nemotron, where Juno offers Whisper.
- Dictation polish Juno has not built: per app tone settings, audio history with export, and a notch overlay.
- It runs on Windows already and iPhone is coming. Juno is Mac only.
- If what you want is dictation, it is the one we would point you at before any paid app on this page.
Where Juno pulls ahead
- Nothing to map. You describe the task in ordinary words and it works out the steps.
- It reads the screen, so it can work software that exposes no Shortcuts, no CLI and no automation of any kind.
- One sentence can cross four apps, including apps that have never heard of each other.
- A wake word, so your hands never have to find the hotkey.
- It runs on macOS 14, where FluidVoice needs 15.
- The CLI hands the same screen, mouse and keyboard to Claude Code, Cursor and Codex over MCP.
- Companion mode watches without clicking, Escape stops it mid action, and every action is logged.
Where the line actually falls
Command Mode and Juno look similar in a feature list and are not the same kind of thing. Command Mode is a router: a phrase you chose, pointed at an action you chose, fired reliably every time. That is genuinely useful and it is faster than an agent for anything you do daily. Juno is for the rest, the jobs you would never bother mapping because you will do them once. Open the invoice in the tab behind this one, pull the totals, and put them in the sheet. There is no line in a config file for that. If you want both, run both, and give them separate keys.
Questions people ask
- FluidVoice already has a Command Mode. Why would I add Juno?
- Because Command Mode runs the commands you set up. If you have not mapped something, it cannot do it, and most people never map more than a handful. Juno needs no mapping because it reads the screen and reasons about what it sees, so it handles the one off jobs that were never worth configuring.
- Is Juno's dictation faster than FluidVoice?
- Probably not. FluidVoice is built around latency and ships models picked for speed, and the site claims under a hundred milliseconds. Juno transcribes locally with Whisper and faster models are what we are working on now. Both are free, so try them against your own voice rather than taking either claim on trust.
- Both are free and both are on GitHub. What is the licence difference?
- FluidVoice is GPLv3. Juno is FSL-1.1-MIT, which is source available and converts to MIT two years after each release. GPLv3 is more permissive today and we are not going to blur that.
- Does Juno cost anything to run?
- Dictation is local and free. Agent tasks run on a model you supply, either the Claude plan you already pay for or an API key, so those cost what the model costs. FluidVoice has no model bill at all because it does not run an agent.
- Can I run both?
- Yes, and it is a sensible setup. Keep FluidVoice for fast dictation and mapped commands, and give Juno a separate trigger for the tasks worth handing off. Just make sure they do not share a hotkey.
Free, and the source is public. macOS 14+.
Download Juno